MUM , the Unified Multitasking Model of Google , was one of the main announcements that Google I / O 2021 left us. It is an artificial intelligence whose purpose is to significantly improve search results. Although the company described its operation during the event, now it has finally offered us information on how MUM is helping to improve its search engine in the midst of a pandemic .
Google recognizes that facing the health crisis was an unprecedented challenge for the search engine . The reason? People searched for information about the virus based on its location and, of course, on language. “We had to learn to identify all the different ways of referring to the new coronavirus in order to provide quality and adequate information from reliable health authorities such as the World Health Organization and the Centers for Disease Control and Prevention,” they point out.
Google's MUM comprises hundreds of vaccine names
After the experience they obtained the previous year with the term “Coronavirus” and its many variants, Google was able to apply this learning in a directly related situation: the names of vaccines. According to their data, there are more than 800 ways to name the vaccines depending on the pharmaceutical involved and the aforementioned language. However, this time they had an ally who facilitated the entire process: MUM.
Thanks to the Unified Multitasking Model, Google was able to identify the names to be able to offer recent, reliable and quality information on vaccines . “With MUM we have been able to identify more than 800 variants of vaccine names in more than 50 languages in a matter of seconds ,” they indicated. This is an outstanding advance because, had it required human intervention, several weeks of work would have been required. Time is money when you need to offer reliable information on a topic so important to humanity.
MUM and its ability to transfer knowledge between languages
How does MUM achieve such an outstanding performance? Like any other artificial intelligence: training. However, the model stands out from the rest thanks to a feature that Google describes as “Transfer of knowledge between languages” . Basically, MUM can understand information in one language and share that knowledge with more than 75 languages. What is transferred, in reality, is the understanding of the data, not a simple translation of it.
This is essential because MUM does not have the need to learn the same thing in a different language . “Thus, thanks to its ability to transfer knowledge, MUM does not have to learn a new capacity or skill in each new language, but can transfer what it has learned from one language to another, which helps us to incorporate improvements quickly. and on a large scale, even when there is not much data with which to train the tool “.
The future of searches
Obviously, this is only MUM's first step in a real search. Google's idea is to implement the model in its search engine in the not too distant future . It would be possible, for example, to offer search results based on where you are, the weather and the language, among other parameters. Even more interesting, it can recognize images to understand them together with the text and enrich the results.
“Our early trials indicate that MUM will not only be able to improve many aspects of our current systems, but it will also help us create entirely new ways of finding and analyzing information.”
Finally, it was interesting to know that Google is doing a kind of internet segmentation to make sure that the results offered belong to trusted websites. If a site is classified by biased, false or explicit content, to mention just a few examples, it will obviously not be taken into account. The above is part of the responsibility of MUM of Google when providing results.
